USB CPPI DMA patch for davinci

This patch fixed USB unstability issue while writing
huge amount of data into usb device.

drivers/usb/musb/cppi_dma.h

@@ -11,120 +11,150 @@
 
 #include "musb_dma.h"
 #include "musb_core.h"
+#include "davinci.h"
 
+/* hOptions bit masks for CPPI BDs */
+#define CPPI_SOP_SET	((u32)(1 << 31))
+#define CPPI_EOP_SET	((u32)(1 << 30))
+#define CPPI_OWN_SET	((u32)(1 << 29))	/* owned by cppi */
+#define CPPI_EOQ_MASK	((u32)(1 << 28))
+#define CPPI_ZERO_SET	((u32)(1 << 23))	/* rx saw zlp; tx issues one */
+#define CPPI_RXABT_MASK	((u32)(1 << 19))	/* need more rx buffers */
 
-/* FIXME fully isolate CPPI from DaVinci ... the "CPPI generic" registers
- * would seem to be shared with the TUSB6020 (over VLYNQ).
- */
-
-#include "davinci.h"
+#define CPPI_RECV_PKTLEN_MASK 0xFFFF
+#define CPPI_BUFFER_LEN_MASK 0xFFFF
 
+#define CPPI_TEAR_READY ((u32)(1 << 31))
+#define CPPI_CHNUM_BITS_MASK  0x3
 
 /* CPPI RX/TX state RAM */
 
 struct cppi_tx_stateram {
-	u32 tx_head;			/* "DMA packet" head descriptor */
+	u32 tx_head;		    	/* "DMA packet" head descriptor */
 	u32 tx_buf;
-	u32 tx_current;			/* current descriptor */
+	u32 tx_current;		 	/* current descriptor */
 	u32 tx_buf_current;
-	u32 tx_info;			/* flags, remaining buflen */
+	u32 tx_info;		    	/* flags, remaining buflen */
 	u32 tx_rem_len;
-	u32 tx_dummy;			/* unused */
+	u32 tx_dummy;		   	/* unused */
 	u32 tx_complete;
 };
 
 struct cppi_rx_stateram {
 	u32 rx_skipbytes;
 	u32 rx_head;
-	u32 rx_sop;			/* "DMA packet" head descriptor */
-	u32 rx_current;			/* current descriptor */
+	u32 rx_sop;		     	/* "DMA packet" head descriptor */
+	u32 rx_current;		 	/* current descriptor */
 	u32 rx_buf_current;
 	u32 rx_len_len;
 	u32 rx_cnt_cnt;
 	u32 rx_complete;
 };
 
-/* hw_options bits in CPPI buffer descriptors */
-#define CPPI_SOP_SET	((u32)(1 << 31))
-#define CPPI_EOP_SET	((u32)(1 << 30))
-#define CPPI_OWN_SET	((u32)(1 << 29))	/* owned by cppi */
-#define CPPI_EOQ_MASK	((u32)(1 << 28))
-#define CPPI_ZERO_SET	((u32)(1 << 23))	/* rx saw zlp; tx issues one */
-#define CPPI_RXABT_MASK	((u32)(1 << 19))	/* need more rx buffers */
-
-#define CPPI_RECV_PKTLEN_MASK 0xFFFF
-#define CPPI_BUFFER_LEN_MASK 0xFFFF
-
-#define CPPI_TEAR_READY ((u32)(1 << 31))
-
 /* CPPI data structure definitions */
 
-#define	CPPI_DESCRIPTOR_ALIGN	16	/* bytes; 5-dec docs say 4-byte align */
+/**
+ *  CPPI  Buffer Descriptor
+ *
+ *  Buffer Descriptor structure for USB OTG Module CPPI.Using the same across
+ *  Tx/Rx
+ */
+
+#define	CPPI_DESCRIPTOR_ALIGN	16	/* bytes; 5-dec docs say 4-byte align*/
 
 struct cppi_descriptor {
-	/* hardware overlay */
-	u32		hw_next;	/* next buffer descriptor Pointer */
-	u32		hw_bufp;	/* i/o buffer pointer */
-	u32		hw_off_len;	/* buffer_offset16, buffer_length16 */
-	u32		hw_options;	/* flags:  SOP, EOP etc*/
-
-	struct cppi_descriptor *next;
-	dma_addr_t	dma;		/* address of this descriptor */
-	u32		buflen;		/* for RX: original buffer length */
+	/* Hardware Overlay */
+	u32 hw_next;     /**< Next(hardware) Buffer Descriptor Pointer */
+	u32 hw_bufp;	   /**<Buffer Pointer (dma_addr_t) */
+	u32 hw_off_len;	    /**<Buffer_offset16,buffer_length16 */
+	u32 hw_options;	    /**<Option fields for SOP,EOP etc*/
+
+	struct cppi_descriptor *next; /**<Next(software) Buffer Descriptor
+					*pointer
+					*/
+	dma_addr_t dma;		/* address of this descriptor */
+	/* for Rx Desc, keep track of enqueued Buffer len to detect
+	 * short packets
+	 */
+	u32 buflen;
 } __attribute__ ((aligned(CPPI_DESCRIPTOR_ALIGN)));
 
-
+/* forward declaration for CppiDmaController structure */
 struct cppi;
 
-/* CPPI  Channel Control structure */
+/**
+ *  Channel Control Structure
+ *
+ * CPPI  Channel Control structure. Using he same for Tx/Rx. If need be
+ * derive out of this later.
+ */
 struct cppi_channel {
-	struct dma_channel	channel;
+	/* First field must be dma_channel for easy type casting
+	 * FIXME just use container_of() and be typesafe instead!
+	 */
+	struct dma_channel channel;
 
-	/* back pointer to the DMA controller structure */
-	struct cppi		*controller;
+	/* back pointer to the Dma Controller structure */
+	struct cppi *controller;
 
 	/* which direction of which endpoint? */
-	struct musb_hw_ep	*hw_ep;
-	bool			transmit;
-	u8			index;
+	struct musb_hw_ep *hw_ep;
+	u8 transmit;
+	u8 index;
 
 	/* DMA modes:  RNDIS or "transparent" */
-	u8			is_rndis;
-
+	u8 is_rndis;
+	u8 autoreq;		/* For now keep this remove this
+				 * one RNDIS + length <64 segmenstation
+				 * will done
+				 */
+	/* Rx Requested mode */
+	u8 rxmode;
+	u8 reqcomplete;		/* zero packet handling*/
 	/* book keeping for current transfer request */
-	dma_addr_t		buf_dma;
-	u32			buf_len;
-	u32			maxpacket;
-	u32			offset;		/* dma requested */
+	dma_addr_t buf_dma;
+	u32 buf_len;
+	u32 maxpacket;
+	u32 offset;		/* requested segments */
+	u32 actuallen;		/* completed (Channel.actual) */
 
-	void __iomem		*state_ram;	/* CPPI state */
-
-	struct cppi_descriptor	*freelist;
+	void __iomem *state_ram;	/* CPPI state */
 
 	/* BD management fields */
-	struct cppi_descriptor	*head;
-	struct cppi_descriptor	*tail;
-	struct cppi_descriptor	*last_processed;
+	struct cppi_descriptor *freelist;	/* Free BD Pool head pointer */
+	struct cppi_descriptor *head;
+	struct cppi_descriptor *tail;
+	struct cppi_descriptor *last_processed;
 
 	/* use tx_complete in host role to track endpoints waiting for
 	 * FIFONOTEMPTY to clear.
 	 */
-	struct list_head	tx_complete;
+	struct list_head tx_complete;
+	u32 iso_desc;
 };
 
-/* CPPI DMA controller object */
+/**
+ *  CPPI Dma Controller Object
+ *
+ *  CPPI Dma controller object.Encapsulates all bookeeping and Data
+ *  structures pertaining to the CPPI Dma Controller.
+ */
 struct cppi {
-	struct dma_controller		controller;
-	struct musb			*musb;
-	void __iomem			*mregs;		/* Mentor regs */
-	void __iomem			*tibase;	/* TI/CPPI regs */
+	/* FIXME switchover to container_of() and remove the
+	 * unsafe typecasts...
+	 */
+	struct dma_controller controller;
+	struct musb *musb;
+	void __iomem *mregs;
+	void __iomem *tibase;
+
 
-	struct cppi_channel		tx[MUSB_C_NUM_EPT - 1];
-	struct cppi_channel		rx[MUSB_C_NUM_EPR - 1];
+	struct cppi_channel tx[MUSB_C_NUM_EPT - 1];
+	struct cppi_channel rx[MUSB_C_NUM_EPR - 1];
 
-	struct dma_pool			*pool;
+	struct dma_pool *pool;
 
-	struct list_head		tx_complete;
+	struct list_head tx_complete;
 };
 
 /* irq handling hook */

drivers/usb/musb/musb_core.c

@@ -60,23 +60,23 @@
  *    includes DaVinci EVM in a common non-OTG mode.
  *
  *      * Control and bulk use dedicated endpoints, and there's as
- *        yet no mechanism to either (a) reclaim the hardware when
- *        peripherals are NAKing, which gets complicated with bulk
- *        endpoints, or (b) use more than a single bulk endpoint in
- *        each direction.
+ *	yet no mechanism to either (a) reclaim the hardware when
+ *	peripherals are NAKing, which gets complicated with bulk
+ *	endpoints, or (b) use more than a single bulk endpoint in
+ *	each direction.
  *
- *        RESULT:  one device may be perceived as blocking another one.
+ *	RESULT:  one device may be perceived as blocking another one.
  *
  *      * Interrupt and isochronous will dynamically allocate endpoint
- *        hardware, but (a) there's no record keeping for bandwidth;
- *        (b) in the common case that few endpoints are available, there
- *        is no mechanism to reuse endpoints to talk to multiple devices.
+ *	hardware, but (a) there's no record keeping for bandwidth;
+ *	(b) in the common case that few endpoints are available, there
+ *	is no mechanism to reuse endpoints to talk to multiple devices.
  *
- *        RESULT:  At one extreme, bandwidth can be overcommitted in
- *        some hardware configurations, no faults will be reported.
- *        At the other extreme, the bandwidth capabilities which do
- *        exist tend to be severely undercommitted.  You can't yet hook
- *        up both a keyboard and a mouse to an external USB hub.
+ *	RESULT:  At one extreme, bandwidth can be overcommitted in
+ *	some hardware configurations, no faults will be reported.
+ *	At the other extreme, the bandwidth capabilities which do
+ *	exist tend to be severely undercommitted.  You can't yet hook
+ *	up both a keyboard and a mouse to an external USB hub.
  */
 
 /*
@@ -2028,6 +2028,11 @@ musb_init_controller(struct device *dev, int nIrq, void __iomem *ctrl)
 		musb->xceiv.host = &hcd->self;
 		hcd->power_budget = 2 * (plat->power ? : 250);
 	}
+
+#ifdef CONFIG_ARCH_DAVINCI
+	tasklet_init (&musb->fifo_check, musb_fifo_check_tasklet,
+			(unsigned long)musb);
+#endif
 #endif				/* CONFIG_USB_MUSB_HDRC_HCD */
 
 	/* For the host-only role, we can activate right away.

drivers/usb/musb/musb_core.h

@@ -239,6 +239,12 @@ enum musb_g_ep0_state {
 	(musb_readb((_x)->mregs, MUSB_DEVCTL)&MUSB_DEVCTL_HM)
 
 #define MUSB_MODE(musb) ((musb)->is_host ? "Host" : "Peripheral")
+struct musb_iso_desc {
+	u32     offset;
+	u32     length;
+	u32     status;
+};
+
 
 /******************************** TYPES *************************************/
 
@@ -285,6 +291,9 @@ struct musb_hw_ep {
 
 	u8			rx_reinit;
 	u8			tx_reinit;
+#ifdef CONFIG_ARCH_DAVINCI
+	u8			fifo_flush_check;	 /* Check FIFO empty */
+#endif
 #endif
 
 #ifdef CONFIG_USB_GADGET_MUSB_HDRC
@@ -292,6 +301,9 @@ struct musb_hw_ep {
 	struct musb_ep		ep_in;			/* TX */
 	struct musb_ep		ep_out;			/* RX */
 #endif
+	struct musb_iso_desc *iso_desc;
+	u32     num_iso_desc;
+	u8 zero;
 };
 
 static inline struct usb_request *next_in_request(struct musb_hw_ep *hw_ep)
@@ -343,6 +355,11 @@ struct musb {
 	struct list_head	in_bulk;	/* of musb_qh */
 	struct list_head	out_bulk;	/* of musb_qh */
 	struct musb_qh		*periodic[32];	/* tree of interrupt+iso */
+#ifdef CONFIG_ARCH_DAVINCI
+	struct tasklet_struct	 fifo_check;	/* tasklet for FIFO empty
+						 * status check
+						 */
+#endif
 #endif
 
 	/* called with IRQs blocked; ON/nonzero implies starting a session,

drivers/usb/musb/musb_host.c

@@ -154,7 +154,8 @@ static inline void cppi_host_txdma_start(struct musb_hw_ep *ep)
 
 	/* NOTE: no locks here; caller should lock and select EP */
 	txcsr = musb_readw(ep->regs, MUSB_TXCSR);
-	txcsr |= MUSB_TXCSR_DMAENAB | MUSB_TXCSR_H_WZC_BITS;
+	txcsr |= MUSB_TXCSR_DMAENAB | MUSB_TXCSR_DMAMODE |
+			MUSB_TXCSR_H_WZC_BITS;
 	musb_writew(ep->regs, MUSB_TXCSR, txcsr);
 }
 
@@ -1242,6 +1243,9 @@ void musb_host_tx(struct musb *musb, u8 epnum)
 		/* REVISIT may need to clear FLUSHFIFO ... */
 		musb_writew(epio, MUSB_TXCSR, tx_csr);
 		musb_writeb(epio, MUSB_TXINTERVAL, 0);
+#ifdef CONFIG_ARCH_DAVINCI
+		hw_ep->fifo_flush_check = 0;
+#endif
 
 		done = true;
 	}
@@ -1305,8 +1309,19 @@ void musb_host_tx(struct musb *musb, u8 epnum)
 		/* set status */
 		urb->status = status;
 		urb->actual_length = qh->offset;
+#ifdef CONFIG_ARCH_DAVINCI
+		/* Check for FIFO empty status.  If not wait for the same
+		 * before completing the urb.  This ensures that the toggle
+		 * status is correctly preserved and data will not be lost.
+		 */
+		if ((tx_csr & MUSB_TXCSR_FIFONOTEMPTY) ||
+			(tx_csr & MUSB_TXCSR_TXPKTRDY)) {
+			hw_ep->fifo_flush_check = 1;
+			tasklet_schedule(&musb->fifo_check);
+			goto finish;
+		}
+#endif
 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_OUT);
-
 	} else if (!(tx_csr & MUSB_TXCSR_DMAENAB)) {
 		/* WARN_ON(!buf); */
 
@@ -1358,9 +1373,9 @@ finish:
  *       (a) all URBs terminate with REQPKT cleared and fifo(s) empty;
  *       (b) termination conditions are: short RX, or buffer full;
  *       (c) fault modes include
- *           - iff URB_SHORT_NOT_OK, short RX status is -EREMOTEIO.
- *             (and that endpoint's dma queue stops immediately)
- *           - overflow (full, PLUS more bytes in the terminal packet)
+ *	   - iff URB_SHORT_NOT_OK, short RX status is -EREMOTEIO.
+ *	     (and that endpoint's dma queue stops immediately)
+ *	   - overflow (full, PLUS more bytes in the terminal packet)
  *
  *	So for example, usb-storage sets URB_SHORT_NOT_OK, and would
  *	thus be a great candidate for using mode 1 ... for all but the
@@ -1944,6 +1959,9 @@ static int musb_cleanup_urb(struct urb *urb, struct musb_qh *qh, int is_in)
 		musb_writew(epio, MUSB_TXCSR, csr);
 		/* flush cpu writebuffer */
 		csr = musb_readw(epio, MUSB_TXCSR);
+#ifdef CONFIG_ARCH_DAVINCI
+		ep->fifo_flush_check = 0;
+#endif
 	}
 	if (status == 0)
 		musb_advance_schedule(ep->musb, urb, ep, is_in);
@@ -2139,6 +2157,46 @@ static int musb_bus_suspend(struct usb_hcd *hcd)
 		return 0;
 }
 
+#ifdef CONFIG_ARCH_DAVINCI
+/* Tasklet routine to handle the completion request. Check for Fifo status
+ * before completing the request. Avoids false completions when data is still
+ * in the fifo
+ */
+void musb_fifo_check_tasklet(unsigned long data)
+{
+	struct musb		 *musb = (struct musb *)data;
+	u8			 epnum = 1, sch_tsklt = 0;
+	struct musb_hw_ep	 *hw_ep = NULL;
+	unsigned long		 flags;
+	u16			 csr;
+	struct musb_qh		*qh;
+
+	do {
+		hw_ep = &(musb->endpoints[epnum++]);
+		spin_lock_irqsave(&musb->lock, flags);
+		if (hw_ep->fifo_flush_check) {
+			csr = musb_readw(hw_ep->regs, MUSB_TXCSR);
+			if (((csr & MUSB_TXCSR_FIFONOTEMPTY) ||
+				(csr & MUSB_TXCSR_TXPKTRDY)))
+				sch_tsklt = 1;
+			else {
+				hw_ep->fifo_flush_check = 0;
+				qh = hw_ep->out_qh;
+				musb_advance_schedule(musb, next_urb(qh),
+							hw_ep, USB_DIR_OUT);
+				DBG(6, "Completed Tasklet %d\n", hw_ep->epnum);
+			}
+		}
+
+		spin_unlock_irqrestore(&musb->lock, flags);
+	} while (epnum < MUSB_C_NUM_EPS);
+
+	if (sch_tsklt)
+		tasklet_schedule(&musb->fifo_check);
+}
+#endif
+
+
 static int musb_bus_resume(struct usb_hcd *hcd)
 {
 	/* resuming child port does the work */

drivers/usb/musb/musb_host.h

@@ -88,7 +88,9 @@ extern int musb_hub_status_data(struct usb_hcd *hcd, char *buf);
 extern int musb_hub_control(struct usb_hcd *hcd,
 			u16 typeReq, u16 wValue, u16 wIndex,
 			char *buf, u16 wLength);
-
+#ifdef CONFIG_ARCH_DAVINCI
+extern void musb_fifo_check_tasklet (unsigned long data);
+#endif
 extern const struct hc_driver musb_hc_driver;
 
 static inline struct urb *next_urb(struct musb_qh *qh)
@@ -108,3 +110,4 @@ static inline struct urb *next_urb(struct musb_qh *qh)
 }
 
 #endif				/* _MUSB_HOST_H */
+